The manuscript by Shelton et al. examined the predictive ability of the combinatorial PGx test on citalopram drug levels as compared to single PK genes (CYP2C19, CYP2D6, CYP3A4). The study design is of the highest interest and aims to fill the knowledge gap in the context that most studies are taking the single drug-gene pair approach to examine the impact of genetically-defined enzyme phenotype on drug metabolism. This might be the first manuscript that reports the impact of combined, weight phenotype due to multiple genetic variations on drug metabolism.

Strengths:

1.) The title and abstract are appropriate for the content of the text. The title condenses the paper's context in a few words, contains the key features, and highlights the study aim. The abstract provides an accessible summary of the paper. References are current, relevant, and cited correctly.

2.) The introduction section is well-constructed though the flow of introducing the research topic, indicating the knowledge gap, explaining the importance of current research, stating the study aim. The adequate literature review is embodied in the logical progression from the current knowledge and enough details provided with recent relevant studies. The value of the study along with the clinical significance is fair and corresponding to the knowledge gap brought up after the literature review. A clear study objective is stated as to assess the ability of the combinatorial PGx test to predict medication blood levels.

3.) The study subjects were recruited from the GUIDED trial identified as citalopram/escitalopram users upon screening visit. It seems all subjects were included in the analyses and no major selection bias is identified. Clinical and demographic characteristics are balanced between citalopram and escitalopram users.

4.) Blood concentration: the procedures of blood sampling/processing and concentration quantification are appropriate and should meet the standards. Not controlling the time since the last dose and concomitant medications is an important limitation of study design to point out. Genotyping/Phenotype assignment: subjects were genotyped with GeneSight Psychotropic panel, which supports the validity and reliability of genotyping for this study. The phenotype assignment algorithms are compared and contrasted as a part of the study aims.

5.) Statistical analysis: The use of ANCOVA to compare the drug levels among various genotype/phenotype groups is appropriate. Age and smoking status are appropriate covariates to account for.

6.) Results/discussion: Despite my concerns brought up for the use of analytical methods, the results of statistical results look appropriate with their analyses used. The meaning, relevance, and importance are well-discussed in the discussion section with the focus on the study aims.

I would like to make the following major suggestions:

1.) Statistical analysis and the potential sex effect: I am wondering if sex could affect the results as sex differences in CYP enzymes have been established (CYP3A4 and CYP2C19), related to and/or in addition to genetic variations. I suggest the author provide the rationale of not accounting for sex when examining the impact of genetic factors on citalopram blood levels.

2.) Statistical analysis, terminology: The authors misuse the concept of "multivariate analyses" a couple of times as it is meant the analyses involving multiple dependent variables instead of aim for the multiple predictors (independent variables) included in the statistical models. I suggest the author make changes accordingly (e.g. univariate analyses, group comparisons).

3.) Statistical analysis, Figure 1: As the purpose of this study is to bring additional values by examing the ability of the combined phenotype of multiple CYP genes/enzymes to predict drug levels, the results in Figure 1 seem that only the effect of a single gene is examined. The audience may have questions if you control for the CYP2D6 genotype/phenotype when examing the impact of CYPC19. Please specify.

4.) Statistical analysis, the use of ANCOVA, results in Table 2: To my knowledge, ANCOVA accounts for the potential interaction between predictors (if more than one predictors) automatically. I suggest the author clarifying if the interaction terms (CYP2C19/2D6 x combinatorial PGx test). The author may need to specify. Besides, I am a bit confused about transforming phenotypes and drug-gene interaction into numerous variables. This approach seems to support the purpose of comparing the variability explained by genetic factors, but the author needs to specify whether numerically transformed genetic variables are treated as continuous variables or as ordinal variables ("factored out" in R). I doubt the appropriateness to treat them as continuous variables (-1, 0, 1) as the phenotype variable does not have interval properties (i.e. the change in enzymatic activity of CYP2D6 PM and IM+NM is not the same as the change in IM+NM and UM).

5.) The limitation of not being able to account for co-medication is acknowledged but it is crucial for the drug metabolism PGx study and can highly impact the results. If the medication list for each subject was collected at the baseline visit per protocol, I recommend doing some analyses on co-medications, e.g. the distribution of the inducers and inhibitors of CYP enzymes and other medications that could affect citalopram metabolism among phenotype groups; consider using the concurrent medication as a covariate for association analyses of genetic factors and citalopram level.

I would like to make the following minor suggestions:

1.) Table 1. Baseline Patient Demographics: For race and ethnicity, only the proportion of Hispanic/Latino is specified. I suggest the author provide the proportions of other major race groups (e.g. Caucasians vs African Americans vs Asians). The information about race and ethnicity is important to provide for human-subjects-involved PGx study, which could inform the distribution of CYP genotypes/phenotypes and the external validity to represent the general population.

2.) Results and discussion: Data presentation and interpretations of results are mixed under each subtitle of the results section. I recommend moving data interpretation to the discussion section.

3.) Reference: The results in this study support author's opinion of therapeutic recommendation based CYPC19 only is not sufficient and CYP2D6 and 3A4 should be accounted for. M Jukic, et al.'s study published on Am J Psychiatry in 2018, entitled: "Impact of CYP2C19 Genotype on Escitalopram Exposure and Therapeutic Failure: A Retrospective Study Based on 2,087 Patients" is by far the largest study investigating examining CYP2C19 and escitalopram PK and seems to reveal higher predictive power of CYP2C19 alone as compared to this study. It is missed in the paper's reference list but necessary to consider referring and commenting on in the discussion section in terms of the comparison of the predictive power of CYP2C19 alone and the combinatorial PGx test in citalopram metabolism.

4.) Table S1. It seems the updated CPIC CYP2D6 genotype to phenotype assignment was referred to as Caudle et al., 2019 is cited. The CYP2D6 phenotype with 2 decreased function alleles (not *10) is offered an activity score of 1, considered as intermediate metabolizer in the standardized CPIC system, with no discrepancy with the GeneSight algorithm. Please confirm.